HOW RESPONSIVE ARE PRIVATE TRANSFERS TO INCOME?
EVIDENCE FROM A LAISSEZ-FAIRE ECONOMY
Donald Cox* Bruce E. Hansen** and Emmanuel Jimenez***
November 1999
JEL Classification: O15, O16, J14, I30, H55
Keywords: Altruism, Exchange, Private Transfers, Crowding Out
*Department of Economics, Boston College.
**University of Wisconsin, Madison.
***The World Bank.
This work was supported by RPO 676-24 of the World Bank’s Research Committee Support Budget and
the National Institute on Aging. We would like to thank Richard Arnott and Andrew Foster for valuable
comments on a previous draft. We also wish to thank seminar participants at the RAND Corporation,
University of Southern California and the 1997 Northeast Universities Development Consortium for
useful comments on earlier drafts. Homi Kharas provided us with the FIES data set. A GAUSS program
which replicates the empirical work can be obtained at www.ssc.wisc.edu/~bhansen. Hansen thanks the
National Science Foundation and the Sloan Foundation for research support. The views expressed here
are the authors’ own and should not be attributed to the Government of the Philippines.
ABSTRACT
In recent years there has been rapidly growing interest in the responsiveness of
private transfers to household income, or “transfer derivatives.” Undoubtedly most of the
interest in transfer derivatives is fueled by the problem of “crowding out”: If incomes are
fungible, and private transfers are highly responsive to household resources, expansion of
public transfers could serve merely to supplant private ones, leaving the distribution of
household consumption largely unchanged. Yet there is also an emerging consensus
from empirical work that these transfer derivatives are rather small, at least for the United
States, implying that crowding out might loom larger in the minds of some economists
than in the data.
A possible reason for the lack of evidence for crowding out in a developed
country like the United States is that the substantial public transfers that already exist
could have rendered it a fait accompli, leaving the remaining small samples
uninformative about crowding out. In this paper, we focus on the Philippines, a country
with extremely limited public income redistribution. We also pay attention to the
possibility that theoretical models of private transfers, including models of altruism and
household risk-sharing, can imply a non-linear relationship between inflows of private
transfers and household resources, taking the form of a spline. We estimate this model by
non-linear least squares, treating the threshold (knot point) as an unknown parameter,
using recently developed econometric techniques. This allows for a more exact
measurement of transfer derivatives than the more commonly applied monotonic
approach. We find that private transfers are widespread, and that they can be highly
responsive to household economic status, in a way that is consistent with either altruistic
preferences, effective household risk-sharing, or both. A strong transfer derivative
occurs, however, only for the very poorest households, where decreases in their resources
prompt very large increases in private transfers.
Our findings have significant policy implications, because they imply that
attempts to improve the status of the very poorest households could be thwarted by
private responses. Some of the gains from public transfers would be shared with richer
households whose burden of support for their less fortunate kin is eased. So the problems
that operative private transfers create for public income redistribution, first pointed out by
Becker and Barro nearly 25 years ago, do indeed matter empirically.
I. INTRODUCTION
How responsive are private, inter-household income transfers to household
resources? This relationship, called “transfer derivatives,” is important because it has
both normative and positive implications. If private transfers respond sharply to changes
in household resources, government-mandated transfers could conceivably have limited
distributional effect, because they might simply “crowd out” private ones. From the
earliest treatment of private transfers, in the work of Becker (1974) and Barro (1974),
crowding out has been raised as a theoretical possibility. In addition to policy concerns,
we may be interested in the responsiveness of private transfers to household income
because it may shed light on the underlying motivation for them. For example, the
altruistic model of transfer behavior predicts that private transfers increase significantly
with reductions in household resources. Similarly, if transfers are used to mitigate risks
that households face, we would expect them to follow the same qualitative pattern, even
if they are not altruistically motivated. Alternatively, if they are used as payments in a
two-way exchange, they could be either positively or negatively related to household
resources (Cox [1987]).
The existing evidence, however, suggests that “crowding out” is a non-issue, at
least for developed countries. There are no empirical studies that indicate a dramatic
response of private transfers to household resources. Most imply that a dollar reduction
in the income of transfer recipients would be met with just a few cents of extra transfers,
and some indicate there might even be a small reduction. Cox and Jakubson (1995), for
example, estimate that a dollar increase in public welfare spending would likely be met
with no more than a 12 cent reduction in private transfers. Altonji, Hayashi and Kotlikoff
(1997) find similarly tepid transfer derivatives. Despite the focus in the theoretical
2
literature on the specter of crowding out, there does not seem to be much evidence for it,
generating considerable tension between theory and empirics.
The United States might be the wrong place to look for crowding out, however,
because its existing public spending may have already rendered it a fait accompli. We
can only speculate about the exact magnitude of the reduction in private transfers caused
by the early expansions of public income redistribution in the United States in the 1930’s
and 1960’s, for example, because a consistent time-series for private transfers does not
exist. However, another form of private transfer, charity, appears to have been displaced
dollar-for-dollar by the public relief programs of the 1930’s (Roberts [1984]).1
This reasoning suggests that we turn to other settings to examine private transfers,
which are not nearly as widespread or as large a fraction of income in the United States as
in developing countries. Data from countries with smaller systems of public
redistribution might afford a cleaner test of the crowding out hypothesis.
There is perhaps a second reason for the paucity of evidence for crowding out.
Even if economists look in the right places (i.e., countries with meager public income
redistribution) they might mistakenly mis-measure the strength of transfer derivatives
because of model mis-specification. For example, a variant of the Becker altruist model,
augmented to include the possibility of inter-household exchange, implies that the
transfer-income relationship can be highly non-linear. Other commonly used models of
household transfers, such as the risk-sharing model, in which preferences are not
necessarily altruistic, can also imply non-constant transfer derivatives. In our empirical
1In
addition to private transfers and charity, there are several other possible private behavioral responses to
social insurance: labor supply (Moffitt [1992]), savings (Feldstein [1974]), and private insurance purchases
(Cutler and Gruber [1996]), for example. Cox and Jimenez (1995) argue that, while most behavioral
responses redound to program recipients, private transfer distortions are more troublesome for targeting
poverty alleviation because they tend to benefit high-income private donors.
3
work, we posit that the non-linearities take the form of a spline. We estimate this model
by non-linear least squares, treating the threshold as an unknown parameter, using
recently developed econometric techniques (Hansen [1996], Chan and Tsay [1998]). Our
empirical work indicates that failure to specify such non-linearity can cause transfer
derivatives to be severely mis-measured.
The intuition for the non-linearity in income-transfer relationships comes from
thinking about the strength and type of transfer motive that is likely to prevail depending
on the potential recipient’s income. Imagine a household facing especially dire straits
and receiving outside private help. The transfers it receives might conceivably be
motivated by a friend or relative’s unvarnished altruism, much in the same way that the
rescue of a drowning man might be motivated by concern for the victim’s life and
nothing else. These altruistic transfers are predicted to have especially strong transfer
derivatives (Becker [1974]). But let the household’s own resources expand and the
altruistic motivation for transfers may evaporate. This doesn’t imply that private
transfers necessarily disappear, because they could still occur, only now for non-altruistic
reasons, such as payments in exchange. The key point is that these other transfers need
not have as dramatic transfer derivatives. The transfer derivative depends on the transfer
motive, which in turn depends on household pre-transfer income.
We use the altruism motive only as an illustration. We cannot prove that
transfers, even if they are highly responsive to household incomes, are necessarily
altruistic. For example, perhaps households enter into mutually beneficial, self-interested
co-insurance contracts, which can imply similar non-linearities in transfer derivatives.
Household A might promise to provide help to B once B’s resources drop below a certain
4
level. Contracts like these can generate non-linearities in the transfer income
relationship.
We use data for the Philippines, a country with minimal public income
redistribution, and find that private transfers follow a sharp, non-linear pattern which
implies substantial crowding out for lower-income households, but much weaker transfer
derivatives for other households. We also find that ignoring the non-monotonicity in the
transfer-income relationship generates deceptively small transfer derivatives, similar to
the ones that characterize many studies of transfer behavior which ignore non-linearities.
Our findings have potentially far-reaching implications for government behavior and
policy, implying that government attempts to alter the distribution of economic wellbeing can sometimes be thwarted by private behavioral responses. Simple analyses that
ignore these responses can make public transfer programs appear more effective than they
really are.
The findings also could have implications for our understanding of human nature
and underlying motivations for private transfers, but here we must tread cautiously. Our
finding of strong transfer derivatives for lower-income households is consistent with,
though certainly not proof of, altruistic preferences. It is obviously illogical to accept the
hypothesis of altruism--or any other behavioral hypothesis for that matter, because the
same empirical patterns may be generated by, for example, self-interested co-insurance
contracts, or some other process. Still, we find the consistency with the predictions of the
altruism model noteworthy in light of previous theoretical and empirical work. Altruistic
preferences are often invoked on commonsense grounds.2 But after being taken to the
2For example, Becker and Murphy (1988) argue that "The altruism assumption is supported by the many
sacrifices parents frequently make for children. Parents spend money, time, and effort on children through
child care, expenditures on education and health, gifts, and bequests." (Becker and Murphy [1988, p. 3]).
5
data the altruism model has emerged beleaguered--up until now there has been hardly a
shred of evidence to support it. So it is striking that we find transfer derivatives that are
consistent with altruism, though we hasten to add that altruistic preferences are not
necessary for the patterns we find.
II. NON-MONOTONIC TRANSFER-INCOME RELATIONSHIPS
There are several explanations why the transfer-income relationship may be nonmonotonic. They include: the exchange versus altruistic motives for giving (loan
repayments can be modeled within this framework), and the insurance motive for
transfers.
A. Altruism and Exchange
A leading dichotomy in the literature, for example, is that of altruistic versus
exchange-motivated transfers. Recent papers which allude to these alternate motives
include, for example, Laitner (1997), McGarry and Schoeni (1995) and Ioannides and
Kan (1993). We show that this framework, which has become somewhat standard in the
literature on private transfers, can imply a non-constant transfer derivatives once we
account for the possibility that the two motives may co-exist.
When transfer donors have alternating altruistic and selfish concerns, transfer
receipts can exhibit a non-linear pattern in recipient income, first falling sharply and then
leveling off or even rising. The pattern is determined by the motive prevailing at the
margin, and the fact that altruism is more likely to be operative at the margin when
recipients are in dire straits. Donors who help desperately poor relatives or bystanders
who attempt to save drowning victims, act not in expectation of remuneration but solely
out of altruistic concern. Their contributions have a pronounced inverse relationship with
recipient resources. The introduction of a poverty alleviation program, for example, or
the sudden appearance of a rescue squad, allow donors to cut back sharply on the
provision of help. So when recipients are desperate, transfers are altruistically motivated
6
and they fall sharply with improvements in recipient resources. The model of private
transfers which predominates at this stage is the altruism model of Becker [1974].3
But when recipients are well-off enough, transfers cease to be altruistic at the
margin. The donor may still care about the recipient (and be happy to learn, for example,
of the latter’s recent lottery winnings) without going out of his way to make him better
off. But the lack of an altruistic motive need not imply that no transfers occur. Unlike
the Becker-Barro model, in which the transfer motive ceases to be operative once the
limits of altruism are reached, our approach allows for non-altruistic transfers. For
example, the donor might contribute money to the recipient, but with the expectation of
receiving a quid pro quo.4 Unlike altruistic transfers, non-altruistic transfers need not be
bear a strong, inverse relationship to recipient incomes. (Indeed, the two may be
positively related, as in the case where recipients are paid their hourly wage for providing
time-intensive services.) So recipient resources rise to a certain threshold where the
transfer motive switches from altruism to exchange. At that point the relationship
between recipient income and transfer amounts received can change dramatically. Hence
the importance of the non-linear specification between the two variables.
Consider the following simple model of transfers, a variant of the one in Cox
[1987]. Suppose the utility of the donor, Ud , is given by
Ud = U (Cd , s,V(Cr ,s)),
(1)
where V = recipient’s well-being, Ci , i = d,r denote donor and recipient consumption,
and s denotes "services" that the recipient might provide to the donor.5
Expression (1) embodies both altruism and exchange. The donor cares about the
recipient so that ∂Ud ∂V > 0. But the recipient must be compensated with financial
3 Though the analysis below is expressed in terms of “donors” and “recipients,” analyses of altruism are
usually applied to family behavior, and therefore we could just as well substitute, for example, the term
“parent” for donor and “child” for recipient. The donor-recipient labels are chosen merely for convenience.
4In the Filipino context, the quid pro quo might spring from a form of reciprocity called utang na loob,
literally "debt inside oneself," which compels recipients to repay donors [Lopez, 1991].
5 For simplicity, and without losing anything essential, we assume that (1) is additively separable.
7
transfers for any services he provides, because we assume that ∂V ∂s < 0. The partial
effects of the other arguments in (1) are positive. In particular, the donor enjoys services
provided by the recipient. The budget constraints for donor and recipient are Cd = Id − T
and Cr = Ir + T, where T denotes financial transfers given by the donor to the recipient
and the Ii are pre-transfer incomes.
The name "services" is a catch-all term that stands for anything that the recipient
provides to the donor in exchange for financial transfers. Services might include help
with home production or behaving in a way the donor prefers. Alternatively, "services"
might capture financial transfers that the recipient makes to the donor at a later date. In
this instance, if T is a loan from donor to recipient, then s denotes the discounted value of
loan repayments.
We assume that there are no market substitutes for transfers or services so that
donor and recipient are engaged in a bilateral monopoly problem when they make
exchanges. One way to resolve the problem is to assume that the donor dominates the
bargaining arrangement.6 This assumption implies that exchange-motivated transfers
provide exact compensation for services. The "donor-dominates" assumption has the
virtue of nesting the traditional altruism formulation of Becker [1974] and Barro [1974]
in a model that contains both altruistic and exchange motives for transfers.
When is a transfer altruistic and when is it exchange-related? The recipients’
utility from severing relations with the donor is V0 = V (I r, 0)--the recipient receives no
transfers and provides no services. The dominant donor faces the participation constraint
that V = V0 . If the constraint is binding, the transfer is exchange-related--transfers
provide exact compensation for services. Otherwise, transfers are altruistic because they
increase the well-being of the recipient. Suppose transfers are altruistically motivated.
6One alternative is Nash bargaining (e.g., McElroy, 1990) which has the attractive feature of conferring to
the recipient some gains from exchange. But Nash bargaining does not include the Becker-Barro version of
altruism as a special case. Since it is this form of altruism which is of primary interest to us, we opt for the
assumption that the donor is dominant.
8
In this case, ∂T ∂Ir < 0 . Recipients with higher pre-transfer income require smaller
transfers to attain the level of consumption that is optimal from the donor’s point of view.
The effect of Ir on T can be large. For example, with Cobb-Douglas preferences and
equal weighting of donor and recipient utility ( ∂U ∂V = 1 ), the value of ∂T ∂Ir is minus
one-half.
If transfers are exchange motivated, the relationship between T and Ir is
different. Think of transfers as the product of an implicit price of services, p , and s . It
is easy to show that ∂s ∂Ir < 0 and ∂p ∂Ir > 0 (Cox [1987]). So transfers can rise or fall
with Ir depending on whether the price effect dominates the quantity effect. It can be
shown that, if transfers both rise and fall with Ir , they will first rise and then fall,
generating an inverted-U-shaped relationship between transfers and recipient income.7
The participation constraint makes it clear why altruism prevails when recipient
income is low, because in this case the constraint does not bind. Donors with an indigent
relative make transfers strictly to raise his well-being. Once recipients become well of
enough, exchange becomes the operative motive at the margin.
With these results in mind, we describe the relationship between recipient income
and transfers, which is depicted in Figure 1. Start with a very low value for recipient
income--somewhere between 0 and K . The recipient is poor, so the transfer is altruistic,
and transfers fall with recipient income. But once the threshold K is reached, transfers
become exchange-motivated, where they have an inverted-U-shaped relation to recipient
income. Finally, once the recipient’s income becomes large enough ( Ir" ) he becomes
independent of the donor, and transfers cease altogether.
The simple model is informative for empirical work in two ways. First, it
suggests that specifications of transfer functions should be non-linear in recipient
income. Second, the model can address the "crowding-out" of private transfers by public
ones. Suppose a public income transfer scheme raises every low-income household’s
7
See, for example, Cox (1987).
9
income to K . Then altruistically motivated transfers will disappear, leaving only
exchange-related ones. Further increases in public transfers can actually ’’crowd in"
private transfers if they occur along the segment K -- I’r . Finally, if public transfers
expand enough to give everyone a minimum income of Ir" , private transfers may
disappear altogether.
B. Household Risk-Sharing
While one facet of the literature on private transfers emphasizes the role of
altruism and other motivations for private transfers in the family, another, related
approach analyzes private transfers one of many strategies households use to pool risk.
As Becker (1974) noted in his seminal work on social interactions, operative, altruistic
transfers can imply effective risk-sharing between donor and recipient. But altruism is
not necessary for such co-insurance to occur, and there have been many recent papers
which have analyzed the connection between private interhousehold transfers and the
sharing of risks in a non-altruistic environment. We demonstrate below that these
models, like the altruism-exchange framework, can imply non-constant transfer
derivatives.
Townsend (1994) analyzes and tests risk-sharing behavior for a sample of Indian
farm households. To cope with stochastic shocks which affect its resources, a household
participates in a risk-sharing group, an aggregation of households in which resources are
pooled and shared, leaving it vulnerable only to shocks which impinge on the group at
large.
Analytically, the problem of risk-sharing is nearly identical to that which arises in
a family characterized by altruism, except that decisions are made by a “social planner”
rather than by the family head (see, e.g., Mace [1991). In addition, the risk-sharing
literature has tended to emphasize all forms of coping with risk, including, in addition to
inter-household transfers, other forms of household redistribution, financial
intermediation, and the like. And while the empirical work has tended to focus on the
10
“bottom line” of consumption outcomes, rather than the risk-sharing methods themselves,
the implications for transfers are similar to those predicted by the altruism model. If
private, inter-household transfers are used for dealing with risks associated with shortfalls
in household resources, then transfers receipts and recipient resources should be inversely
related. In addition, under full risk-sharing we would expect that transfers would be more
responsive to income at low levels of recipient income, since a disproportionate number
of these households would have been likely to experience negative income shocks to be
remedied by insurance payments. At very low incomes, the effects of these shocks might
be profound as households struggle to survive, necessitating greater “insurance”
payments. For those who are able to access precautionary savings or formal insurance or
credit channels, negative income shocks are less likely to trigger dramatic private transfer
responses. So with household risk-sharing, as with altruistically motivated transfers,
there are likely to be non-linearities in transfer derivatives.8 Moreover, such a
relationship would be discontinuous depending upon access to formal insurance or capital
markets.
III. EVIDENCE ON TRANSFER DERIVATIVES
Are private transfers responsive to recipient incomes? The existing literature
analyzing private transfers in the United States indicates that the answer is no. The
earliest empirical studies of the connection between private transfers and income
concerned bequests to children, and the predominant pattern was found to be equal
sharing among children (e.g., Menchik [1980]). Later studies, such as Wilhelm (1996),
were better able to control for the characteristics of children but produced essentially the
same result, indicating that private transfers, at least in the form of bequests, were
8 Other, more subtle non-linearities may arise from problems of commitment and implementability of risksharing (e.g., Coate and Ravallion [1993]). Co-insurance schemes for protecting against extreme disasters
may not be sustainable between two self-interested parties who always have the option of defecting from
their informal insurance arrangements. This “implementability constraint” serves to limit transfer payments
to households who have experienced extreme shocks. Considerations like these introduce non-linearities
that differ from those in figure 1. Transfers are only responsive to household income if such income is high
enough that the implementability constraint does not bind. Once the shock becomes so severe that this
constraint binds, transfers no longer expand with further reductions in income (see Coate and Ravallion, pp.
10-11).
11
remarkably insensitive to the incomes of potential recipients. Inter-vivos transfers might
be thought to be more compensatory because they are more likely to be deliberate. But
existing studies point either to positive income effects (e.g., Cox [1987]), or negative
effects that are quite small in magnitude (e.g., McGarry and Schoeni [1995]).9
Transfer derivatives have been used to test various motives for private transfers
(e.g., Cox [1987], Altonji, Hayashi and Kotlikoff [1997]). Altruism places a restriction
on these derivatives: A dollar increase in recipient income, accompanied by a dollar
decline in donor income, should prompt an exact dollar reduction in private transfers.
This restriction was tested and rejected by Cox and Rank [1992] and Altonji, Hayashi and
Kotlikoff [1997]. The former find positive recipient income effects for transfers and the
latter find negative income effects, but in each case altruism’s restrictions are rejected
because the estimated transfer derivatives were quite small.
Perhaps one reason for the small income effects on transfers in the United States
is that public income transfer programs have already crowded out the transfers most
likely to be responsive to income. The evidence for developing countries is less clear cut.
Some studies find an inverse relationship between recipient income and transfer amounts
(e.g., Kaufmann and Lindauer [1986; El Salvador], Kaufmann [1982; the Philippines]
and Ravallion and Dearden [1988; rural households in Java]. But other studies find a
positive relationship (e.g., Lucas and Stark [1985; Botswana] Ravallion and Dearden
[1988; urban households in Java] and Cox and Jimenez [1997; urban households in Peru].
A problem with each of these studies, however, is that none contains a specification for
recipient income that is flexible enough to capture the complete transfer pattern depicted
in Figure 1.10
9
For a review of studies which address the separate issue of the overall size of private transfers in the
United States, see Gale and Scholtz (1994).
10The only study which approaches the subject of non-linearities, and finds substantial income effects, is
that of Kaufmann [1982], which we discuss in more detail below. In addition, a recent paper by Schoeni
(1997) emphasizes non-constant transfer derivatives and finds income effects that are qualitatively
consistent with Figure 1. But his estimated transfer derivatives, like those found in the rest of the United
States studies, are minuscule.
12
In sum, lack of flexibility in functional form and the existence of a large welfare
state in the United States may have made it difficult to detect evidence for substantial
transfer derivatives. Since our work is for a country with little in the way of public safety
nets, we should be able to better detect the full range of behavior depicted in Figure 1.
Before proceeding to the empirical work, we place the public safety nets of the
Philippines in perspective.
IV. SOCIAL INSURANCE IN THE PHILIPPINES
How does the Philippines compare to other countries in terms in terms of
allocation of GNP to social welfare spending? The table below indicates rankings for an
assortment of countries for welfare and other payments11. The fraction of GNP devoted to
social welfare spending in the Philippines is several orders of magnitude smaller that of
developed countries such as the United States and the United Kingdom, and it is dwarfed
by those of welfare states like Sweden and the Netherlands. Even among other
developing countries similarly classified as "middle income," such as Morocco, Korea
and Thailand, the share of welfare payments is relatively small.
1988 Public Welfare and Housing Payments as a Percentage of:
Country
GNP
All Government Expenditures
Sweden
22.1
Netherlands
22.1
UK
11.6
USA
7.2
Sri Lanka
3.7
Morocco
2.1
Korea
1.3
Thailand
0.1
Indonesia
0.04
Philippines
0.03
Uganda
0.03
_____________________________________
11These
54.2
39.6
30.9
31.5
11.7
7.3
8.5
5.4
1.7
2.2
2.9
Per Capita GNP
(1988 US $)
19,300
14,520
12,810
19,840
420
830
3,600
1,000
440
630
280
numbers include a wider range of categories than would normally be included in "welfare
payments." They cover expenditures on housing and slum clearance activities, sanitary services,
compensation payments for sickness and disability, pensions, unemployment payments, family and child
allowances, and the cost of providing welfare services such as care for the elderly, the disabled and children.
Nonetheless, they are indicative of the relative importance of welfare payments across countries.
13
Source: World Development Report 1990, "World Development Indicators," Tables 1 and 11, World Bank,
Washington, DC 1990.
The main components of the public safety net in the Philippines are food subsidy
programs, public works employment, livelihood creation (i.e., training) programs, credit
for small and medium enterprises and farms. There is a formal social security program
that provides benefits for old-age, disability, death, workman’s injury, illness and
maternity. But this system only covers government workers and 60 percent of workers in
the formal sector.
According to a recent World Bank evaluation, welfare spending in the Philippines
is costly to administer and poorly targeted:
Few of the resources currently spent on government livelihood programs or
labor-based public works reach the poorest groups. Even the cash and in-kind
transfer programs do not target effectively. (p. 49, World Bank [1996]).
The same report also cautioned that "The public provision of a safety net for the poorest
must be designed to complement rather than replace existing private arrangements of
private transfers, which are already widespread in the Philippines (p. 44)." However,
many discussions of private transfer behavior are necessarily impressionistic, because,
until recently, information about such transfers was not commonly available in the
Philippines.
V. EMPIRICAL WORK
A. Descriptive Overview
The Family Income and Expenditures Survey (FIES) of the Philippines is
undertaken every three years to gather income and consumption information for a
representative cross-section of Filipino households. The main objective of the survey is
to obtain information about expenditure patterns, income sources and inequality and to
14
obtain information for updating weights in the consumer price index. The FIES is the
only official household survey of income and spending patterns which is nationally
representative.12
The survey gathers information about sources of income (both cash and in-kind),
demographic variables such as family size, marital status and number of children by age
group, and job-related information such as earnings and employment status. In addition,
the survey asks respondents to report on a variety of transfers, both in-kind and cash,
from domestic sources and from overseas. The survey’s definition of private transfers
includes only interhousehold transfers, so that redistribution within the household is not
measured. The FIES covers 18,922 households: 8,863 urban households and 10,059
rural households.
Tables 1 and 2 list household characteristics according to whether households
were net recipients or givers of transfers. Table 1 refers to urban households and Table 2
to rural ones. We consider the two separately because of the large differences between
urban and rural standards of living.
Nearly all households participate in transfer networks, and private transfers are
large component of total household income. Eighty-eight percent of the urban
households were involved with transfers, either as donors, recipients, or both. In the
urban sample gross transfer receipts accounted for 12 percent of total household income.
Among urban recipients, these receipts accounted for nearly 20 percent of total household
income.
More households received transfers than gave them, in part because international
remittances account for a large fraction of total receipts. Another reason for the
discrepancy is that households were asked more questions about receipts than gifts. Only
one module in the survey was concerned with gifts and respondents were asked only
12There is no public use file that is as easily available as similar data sets in the United States. We were
able to get a copy of the 1988 FIES as part of a team working on a World Bank basic economic report, and a
copies of the data files that we use are available upon request from the authors.
15
summary questions. Three modules dealt with receipts, including the value of in-kind
transfers and those received from abroad. Therefore, gifts are likely to be under-reported
relative to receipts.13 Despite this, an examination of household characteristics according
to their private transfer behavior conveys useful information about private transfer flows.
Consider the urban households in Table 1. Recipients have the lowest average pretransfer income (Table 1, column 1), and donors have the highest. This pattern suggests
that transfers flow from high- to low-income households. The same pattern exists for
rural households (Table 2). Ninety-three percent of them were involved with private
transfers and gross transfer receipts made up 12 percent of total income. In both samples,
transfer recipients are less educated than donors, and the proportions of female-headed
households and households whose head is not employed are higher among recipients than
donors.
Private transfers help equalize the distribution of income. In previous work using
the data set we found that, for the sample overall, private transfers increase the income
share of the lowest quintile by 62 percent, and reduce the share of the top quintile by
nearly 6 percent [Cox and Jimenez, 1995]. We also found that private transfers appeared
to alleviate poverty, in the sense that poverty rates calculated without including private
transfers were much higher than actual ones.14 But we did not use a rigorous framework
for understanding or estimating non-linearities in the income-transfer relationship.
B. Estimates of Transfer Functions
How responsive are private transfers to income? To answer this question, we
estimate a regression spline model for net transfer receipts, T , defined as gross transfers
received minus gross transfers given.
1) Specification
13In
the usual case we might suspect under-reporting to work in the opposite direction to the extent that
donors exaggerate their generosity or recipients downplay their dependency.
14Urban poverty rates were one-third higher without private transfers--42 versus 32 percent. The
comparable figures for rural households were 66 versus 59 percent (Cox and Jimenez [1995]).
16
To model nonlinearity, we use a continuous linear spline with a single knot. A
continuous linear spline is similar to a threshold regression, with the important exception
that the regression function is constrained to be continuous in pre-transfer income I . Let
K denote the knot, or threshold level of income (as in Figure 1) at which transfer
behavior switches from altruistic to non-altruistic. Let d1 (K) denote a dummy variable
which takes the value 1 if I ≤ K and 0 if I > K , and let d2 (K) = 1 − d(k) denote a
dummy variable which takes the value 1 if I > K . For fixed K the continuous linear
spline is a linear function of the variables (I − K)* d1 (k) and (I − K)* d2 (K) as well as
the other regressors including an intercept.
If the knot K were known, the model could be estimated by ordinary least squares
(OLS). Since K is unknown, it should be estimated along with the other regression
parameters.15 Non-linear least squares (NLLS) is the appropriate estimation criteria.
Since the knot enters in a non-linear and non-differential manner, conventional gradient
search techniques to implement NLLS are inappropriate. Instead, we employ a method
which is sometimes called conditional least squares, and works as follows. For any K
the model is estimated by OLS, yielding the sum of squared errors as a function of K .
The least squares estimate of K is found by searching over K and selecting the value
which yields the lowest sum of squared errors.16 For this value of K the slope
parameters are estimated by OLS. Chan and Tsay (1998) have shown that these NLLS
estimates are consistent and asymptotically normal, and these authors also provide
estimates of the asymptotic variance-covariance matrix.
To reduce the influence of the extremely wealthy, we omitted the top 2%
wealthiest households (measured by pre-transfer income) from each sample. Since we
15In earlier, descriptive work [Cox and Jimenez, 1995] we estimated a spline function for transfer receipts
but treated the non-linearity in an ad hoc way by fixing the nodes of the spline at quartiles in pre-transfer
income.
16For a sample of size n with m parameters, at most n-2m distinct values for K may be considered. For a
complete search, it is sufficient to examine the values of K equaling the sample values of the threshold
variable (in our case I), excluding the extremely smallest and largest values to ensure that at least m
observations lie on each side of the threshold. This is what we did for our calculations.
17
are treating pre-transfer income as exogenous, this does not induce sample selection bias.
As a regressor, we included a dummy variable for households with zero pre-transfer
income. (The sensitivity of the results to this specification are discussed later.) We also
include the level of retirement income and a dummy for the presence of retirement
income, to account for the differential behavior of retirees.
We include other household characteristics in addition to pre-transfer income in
the transfer function. Education of the household head is included to capture permanent
income effects. Age is included because recent evidence indicates that liquidity
constraints may play an important role in private transfer behavior (e.g., Cox [1990],
Guiso and Jappelli [1991], Cox and Jimenez [1997]). Binding liquidity constraints would
affect the timing of transfers. For example, they may be targeted to younger households
who have not yet established their reputations in credit markets. We include a dummy
variable indicating whether the household is headed by a woman because nearly all
studies of private transfers indicate that they are disproportionately targeted toward
women (e.g., Lucas and Stark [1985 ,Botswana]; Kaufmann and Lindauer [1986 ,El
Salvador]; Cox and Jimenez [1997, Peru]; Guiso and Jappelli [1991, Italy]; Cox [1987,
United States]). Further we include a dummy for whether the household head is
employed and whether husband and wife are both employed. In addition, we control for
marital status, household size and composition.
Female-headed and married households receive much more transfers than others.
The gender effect is consistent with the long list of other studies of private transfers that
find that transfers are disproportionately targeted toward women. Evidence for marriage
effects for transfers is more ambiguous; some studies find positive effects and others
negative ones. But part of these demographic effects could be the result of overseas
remittances. Consider a household in which the primary earner is the husband. Suppose
that he is currently working abroad and remitting part of his income to his wife and
children. Since transfers from abroad are much larger than those from domestic sources
18
(Table 1), especially large transfers would tend to accrue to married, female-headed
households.17 We control for this situation by including a dummy for these households.18
2) Estimated Transfer Functions
Transfer function estimates and standard errors19 for urban households are given
in the first two columns of Table 3. The key parameters are the first three. The estimate
of the knot (income threshold) is 20,080 pesos, which corresponds to the 25th sample
income quantile. For incomes below the 25th percentile, the slope of the transfer function
is -0.37, implying a high trade-off between income sources. For incomes above the
threshold, the estimated transfer function is utterly flat. A good understanding of the
transfer function can be found by examining Figure 2, which plots the estimated transfer
function (holding all other regressors fixed at their sample means). The non-linear effect
of income on transfers is striking and strong, and is consistent with the theory of section
II. The point estimates are consistent with the idea, for example, that altruistic transfers
are important for low-income households, but are not operative for higher-income
groups.20 On a final note, households with no pre-transfer income receive an additional
11,586 pesos in transfers above that suggested by the transfer function displayed in
Figure 2. This suggests a very strong non-linearity for very low incomes.
It is revealing to contrast the spline with a standard linear specification. If the
spline is replaced by a standard linear function, the OLS estimate of the coefficient on
income drops to the tiny value -0.014. Thus a researcher who tests the effect of income
17Among those urban households who receive a transfer from abroad, the proportion of married female
heads is over 3 times higher than in the urban sample overall (5 versus 17 percent).
18 This takes into account a major concern that the motives of non-residents who are still considered
household members may be different from those of others. What we cannot control for is the possibility
that the female may be absent.
19 The reported standard errors are appropriate for homoskedastic errors. Standard errors robust to
heteroskedasticity were also calculated, but were not meaningfully different.
20The finding of such strong recipient income effects is all the more striking in light of the fact that we
cannot include the income of the donor household in the regression. The omission of donor income is likely
to bias recipient income coefficients toward zero. Evidence from the United States (Cox and Rank [1992],
Altonji, Hayashi and Kotlikoff [1997] and for Peru (Cox and Jimenez [1995]) indicates that such omitted
variable bias is indeed positive, though numerically small.
19
on transfers by fitting a linear model will erroneously conclude that there is no
meaningful relationship.
The estimates and standard errors for rural households appear in the third and
fourth columns of Table 3, and a plot of the estimated transfer function is given in Figure
3. Rural patterns for transfers are quite similar to urban ones. The most striking
similarity is that for income levels below the threshold, the transfer function has the
identical slope of -0.37. A slight difference is that for income levels above the threshold,
rural households have a slope of –0.03, which is significantly different than zero.
Another difference is that the income threshold (to switch from altruistic to non-altruistic
transfers) is 10,576 pesos, which is about one-half that of the urban sample. Since rural
incomes are lower, however, the threshold of 10,576 pesos corresponds to the 20th
income quantile for the rural sample, which is a point in the income distribution similar to
that found for the urban sample.
We assessed the fit of the linear spline through a series of specification tests.
First, we replaced the linear spline in income I with an eighth-order polynomial. The
estimated transfer functions are displayed in Figures 2 and 3 along with the linear spline
estimates. The polynomial functions closely match the shape of the linear spline, giving
strong informal support for our chosen specification. (In the following section we discuss
formal tests for the spline specification against the polynomial alternative.) Second, we
omitted the dummy for households with zero pre-transfer income. For the urban sample,
the slope of the transfer function for altruistic transfers (low pre-transfer income)
steepened to –0.45, and that for the rural sample steepened to –0.41. Third, we allowed
the spline function to be quadratic for the non-altruistic (high income) regime, but the
quadratic term was statistically insignificant, and the graphs of the estimated transfer
functions were not meaningfully different than the graphs for the linear spline presented
in Figures 2 and 3. Fourth, we included a quadratic term in the age of the household
head, but this variable was insignificant in both samples.
20
As a final specification test, we estimated a model which allowed for the kink K
to be random across households. Assuming that K is normally distributed with mean κ
and variance ω and independent of the regression error, it is straightforward to calculate
the conditional mean of T given I and the other regressors. By allowing ω>0, this
model softens the kink in the mean transfer function, while ω=0 specializes to the linear
spline. We estimated the parameters by NLLS. Our estimates for both the urban and
rural samples yielded point estimates for ω of 0, which means that the linear spline fits
better than the random knot model. We conclude that the data strongly support the linear
spline specification.
The threshold has another potentially interesting interpretation as the “private”
analog of the Philippine poverty line, which is calculated according to the value of the
assumed minimum consumption bundle for an acceptable standard of living. To the
extent that the node marks the point that triggers changes in household behavior, it could
be interpreted as a poverty line that is perceived by households, either for altruistic
reasons or to compensate for lack of access to formal insurance or credit markets. If so, it
would be interesting to compare to the Philippine poverty line. In 1988, the poverty line
(adjusted to take into account food requirements for the poor) was computed to be 3,800
pesos per person per year (World Bank [1993]). The income threshold we estimate for
urban areas is about 20,080 pesos per household is almost exactly the same, since at 5.3
persons per household, this translates into 3,789 pesos per year.
3) Tests of the Generalized Transfers Model
The generalized theory of private transfers in section II had the strong implication
that the transfer function takes the form of a spline, and this motivated the econometric
specifications reported in Table 3. Since altruism cum exchange implies the spline form,
we can test this hypothesis by testing whether a spline exists in the transfer function.
Testing for the presence of a spline relationship is non-standard from the
statistical point of view. The null hypothesis of no spline is equivalent to that the
21
regression slopes in the two regimes are equal. But in this case, the knot K is not
identified, and it is known that test statistics (such as t- or F-statistics) do not have
conventional asymptotic distributions. Hansen [1996] shows that the bootstrap allows the
calculation of asymptotically valid p-values in this context. Technical details are given in
the appendix.
We report bootstrap p-values for the two samples in Table 4. For each model,
1000 bootstrap replications were made. The first line reports Wald statistics and
bootstrap p-values of the test of a linear specification against the alternative of a linear
spline. In neither sample did a single bootstrap statistic exceed the actual value, yielding
estimated p-values of 0.00, which means that linearity is easily rejected in favor of a
linear spline at all significance levels. Similarly, the second line in Table 4 reports tests
of a quadratic transfer function (that is, a linear regression of T on I , I 2 , and the other
control variables) against the alternative of a quadratic spline (the use of a quadratic
spline under the alternative is done so that the hypotheses are nested). Again, the
quadratic specification is rejected at all significance levels for both samples in favor of
the spline specification. The tests show that the spline is necessary to capture the correct
transfer function.
One might ask whether a higher order polynomial could adequately approximate
the unknown transfer function. We repeated our hypothesis tests for transfer functions
modeled as simple polynomials up to order 7. The results are quite interesting. For the
urban samples the spline effect is statistically significant for polynomials of order 5 and
below, which is strong evidence that the transfer effect of altruism prevails for lower
income households. The evidence from the rural samples is less strong, as the spline
effect is marginal in a third order (cubic) polynomial, and insignificant against higherorder polynomials.
Another question is whether the specification of a single knot spline function is
appropriate. To assess this question, we performed a two-step Wald test for a second
22
knot or threshold effect. Since this testing problem is non-standard, we again used
bootstrap methods to assess statistical significance. The results are presented in Table 5.
Neither test statistic is close to significant. This means that with high confidence, the
hypothesis of a linear spline with a single knot cannot be rejected against the alternative
of a linear spline with two knots.
In summary, the evidence shows that there is strong statistical evidence in favor of
a strong kink in the transfer function which is well approximated by a linear spline with a
single knot, and cannot be well approximated by a low-order polynomial. This statistical
evidence supports the generalized model of altruism described above.
4) Other Results
Better educated households receive more transfers, which is consistent with
altruism in the presence of liquidity constraints [Cox, 1990]. When a recipient faces a
binding borrowing constraint, transfers help fill the gap between desired consumption and
permanent income. Suppose that the recipient’s future income rises, with current income
constant. If desired consumption is determined by permanent income, the gap between it
and current income widens. Hence the optimal transfer increases.21
One instance in which urban and rural samples differ is with respect to estimated
age effects. Age-transfer profiles slope downward for urban households but upward for
rural ones. While the effect is not pronounced, the sign pattern does suggest that transfers
are more likely to function as old-age support in rural areas, where pension coverage is
much lower. Negative age effects are also consistent with the idea that private transfers
respond to liquidity constraints, since we would expect that transfers would be directed
toward consumers who have not yet established reputations in formal credit markets.
21Consider
the following simple numerical example. Suppose that an altruistic donor overlaps with a
recipient for two periods. The donor, but not the recipient, has access to capital markets and income of
[200, 200]. The recipient has income of [60, 100]. Assume that the subjective rate of time preference
equals the interest rate and that the donor weights the recipient’s utility equal to his own. The consumption
profile for each person is [140, 140], and first-period transfers are 80. Now suppose that the recipient’s
second-period income increases by 4. His optimal consumption is now 141. With current income constant,
transfers must rise by 1 to enable the recipient to attain this new level.
23
The size and composition of the household appears to matter. The presence of
infants has no measurable effect on transfers, and young children (up to age 7) have no
effect in the rural sample and have a negative effect in the urban sample. Older children
(age 8 through 15) have a slight positive effect, and the number of adults has a larger
positive effect (850 peso per adult for urban, 558 peso for rural).
Households headed by someone who is not employed receive an 8,398 peso boost
(urban) in transfers (5,632 peso for a rural household). On the other hand, if both the
husband and wife both work, then transfer amounts are reduced (1,399 peso for urban,
506 peso for rural). With income constant, dual earner status implies lower earnings per
person, which might prompt altruistically motivated transfers. On the other hand, if the
husband and wife both work, they may be better able to cope with income variability,
which would be consistent with receiving less private transfers.
Finally, we observe that marriage has no effect on transfers, and female-headed
households receive a statistically insignificant boost. An important effect is reserved for
married female-headed households, which receive increased transfers of 32,131 peso in
the urban sample and 15,836 in the rural sample. This is consistent with the hypothesis
that the husbands are working abroad and are remitting their income to their spouses.
VI. CONCLUSION
A primary reason for the surge of interest in family behavior in economics is its
implications for government redistribution policy. Researchers and policymakers are
both becoming aware that the effects of public income redistribution depend in part on
private responses to them, including private transfers. Ironically, it may be that the best
place to investigate the effects of the "crowding out" of private transfers for public ones is
where little public transfers exist, because in these settings it is likely to be easier to
measure the full range of responsive and non-responsive transfers.
24
Our results suggest that the potential for crowding out is indeed large for public
transfers targeted toward those in the lower reaches of the income distribution. Our
spline estimates indicate that, depending on the sample, anywhere from 30 to 80 percent
of private transfers could be crowded out for those in (roughly) the lowest quintile of preprivate transfer income.
Our estimated tradeoff between private transfers and recipient income is much
higher than existing studies, either for the United States or for developing countries, with
one interesting exception that we mentioned earlier. Kaufmann (1982) examined private
transfer behavior among urban households in the Philippines and El Salvador. Kaufmann
develops a theory of transfers that is different from ours, but is nonetheless focused on a
non-linear relationship between transfers and income.22 Accordingly, he investigates
splined and other non-linear specifications of transfer functions, using small household
micro-data sets collected in conjunction with a World Bank project on informal housing.
For the Philippines, he finds that "The large and highly significant income coefficient ...
indicate(s) that the slope of the function is rather steep--and negative--at low income
levels, and that the slope rapidly decreases as income increases." [Kaufmann, p. 172.]
Kaufmann’s estimated tradeoff between transfers and income for low-income households,
-0.625 of a peso, is well within the range of our estimates. And like us, he finds that a
simple linear specification performs poorly relative to non-linear ones.
We think this evidence adds to a compelling case paying close attention to nonlinearities in analyzing private transfer behavior. The spline specification implied by the
theory necessitates a search for the knot point in the transfer-income relationship. Such
searches introduce inference problems, but these can be addressed as shown by Hansen
[1996, 1999]. Future work on private transfers should focus on sharp non-linear
relationships, preferably in settings where public transfers are small.
22Kaufmann
postulates that private transfers targeted to a household’s "basic needs" (e.g., basic food, health
and shelter requirements) respond differently than other transfers.
25
Appendix: The Bootstrap Test for a Spline
The Wald statistic for testing the hypothesis of a polynomial transfer function
against the alternative hypothesis of a polynomial spline function (of the same order) is
W = n(S0-S1)/S1, where S0 is the sum of squared errors for the polynomial null model
(calculated by OLS) and S1 is the sum of squared errors for the spline model . To
compute the bootstrap p-value, generate a sample of iid random variables drawn from the
empirical distribution of the NLLS residuals, and repeat the above procedure, treating the
generated data as the dependent variable. Store the resulting Wald statistics, and
calculate the percentage which are larger than the actual Wald statistic in the sample.
This is the bootstrap p-value.
Since calculation of the spline model involves a large number of regressions, each
of which involves nearly 10,000 observations, we restricted the class of possible
thresholds to 100 values consisting of equaling spaced quantiles between the 5th and 95th
quantiles.
26
REFERENCES
Altonji, Joseph G., Hayashi, Fumio and Kotlikoff, Laurence. "Parental Altruism and Inter
Vivos Transfers: Theory and Evidence." Journal of Political Economy 105 (December
1997): 1121-1166.
Barro, Robert J. "Are Government Bonds Net Wealth?" Journal of Political Economy 82
(November/December, 1974): 1095-1117.
Becker, Gary S. "A Theory of Social Interactions." Journal of Political Economy 82
(November/December, 1974): 1063-94.
__________ and Murphy, Kevin M. "The Family and the State." Journal of Law and
Economics 31 (April 1988a): 1-18.
Chan, K.S. and Ruey S. Tsay. “Limiting Properties of the Least Squares Estimator of a
Continuous Threshold Autoregressive Model,” Biometrika 45 (1998): 413-426.
Coate, Stephen and Ravallion, Martin. “Reciprocity without Commitment: Characterization
and Performance of Informal Insurance Arrangements.” Journal of Development
Economics 40 (February 1993): 1-24.
Cox, Donald. "Motives for Private Income Transfers." Journal of Political Economy 95 (June
1987): 1045-76.
____. "Intergenerational Transfers and Liquidity Constraints." Quarterly Journal of
Economics 105 (February 1990): 187-217.
__________ and Rank, Mark. "Inter-Vivos Transfers and Intergenerational Exchange."
Review of Economics and Statistics 74 (May 1992): 305-314.
__________ and Jakubson, George. “The Connection Between Public Transfers and Private
Interfamily Transfers.” Journal of Public Economics 57 (May 1995): 129-167
__________ and Jimenez, Emmanuel. "Private Transfers and the Effectiveness of Public
Income Redistribution in the Philippines." In Public Spending and the Poor: Theory
and Evidence edited by Dominique van de Walle and Kimberly Nead. Baltimore: The
Johns Hopkins University Press (for The World Bank), 1995.
____ , Eser, Zekeriya, and Jimenez, Emmanuel. "Motives for Private Income Transfers
Over the Life-Cycle: An Analytical Framework and Evidence for Peru." Journal of
Development Economics 55 (February 1998): 57-80.
Cutler, David M. and Gruber, Jonathan. "Does Public Insurance Crowd Out Private
Insurance?" Quarterly Journal of Economics 111 (May 1996): 391-430.
Dunn, Thomas. "The Distribution of Intergenerational Income Transfers Across and Within
Families." Mimeographed. Syracuse, NY: Syracuse University, 1994.
27
Feldstein, Martin. "Social Security, Induced Retirement, and Aggregate Capital
Accumulation." Journal of Political Economy 82 (October, 1974): 905-926.
Gale, William and Scholz, John Karl. "Intergenerational Transfers and the Accumulation of
Wealth." Journal of Economic Perspectives 8 (Fall 1994): 145-160.
Guiso, Luigi and Tullio Jappelli. "Intergenerational Transfers and Capital Market
Imperfections: Evidence from a Cross-Section of Italian Households." European
Economic Review 35 (January 1991): 103-120.
Hansen, Bruce E. "Inference When a Nuisance Parameter is Not Identified Under the Null
Hypothesis." Econometrica 64 (March 1996): 413-430.
Hansen, Bruce E. "Sample Splitting and Threshold Estimation." Econometrica 67 (1999):
forthcoming.
Ioannides, Yannis M. and Kan, Kamhon. “The Nature of Two-Directional Intergenerational
Transfers of Money and Time: An Empirical Analysis.” Manuscript. Blacksburg:
Virginia Polytechnic Institute and State University, March 1993.
Kaufmann, Daniel. "Social Interaction as a Strategy for Survival Among the Urban Poor:
A Theory and Evidence." Ph.D. Dissertation, Harvard University, 1982.
and David Lindauer. "A Model of Income Transfers for the Urban Poor." Journal
of Development Economics 22 (July/August 1986): 337-50.
Laitner, John. “Intergenerational and Interhousehold Economic Links.” in Mark R.
Rosenzweig and Oded Stark, eds., Handbook of Population and Family Economics.
Amsterdam: North-Holland, 1997.
Lopez, Maria Elena. "The Filipino Family as Home for the Aged." Research Report No. 91-7,
Population Studies Center, University of Michigan, January, 1991.
Lucas, Robert E. B. and Oded Stark. "Motivations to Remit: Evidence from Botswana."
Journal of Political Economy 93 (October 1985): 901-18.
Mace, Barbara. “Full Insurance in the Presence of Aggregate Uncertainty.” Journal of
Political Economy 99 (October 1991): 928-956.
McElroy, Marjorie B. "The Empirical Content of Nash-Bargained Behavior." Journal of
Human Resources 25 (Fall 1990): 559-83.
McGarry, Kathleen and Schoeni, Robert F. "Transfer Behavior in the Health and Retirement
Study: Measurement and the Redistribution of Resources Within the Family." Journal
of Human Resources 30 (Supplement 1995): S184-S226.
28
Menchik, Paul L. "Primogeniture, Equal Sharing and the U.S. Distribution of Wealth."
Quarterly Journal of Economics 94 (March 1980): 299-316.
Moffitt, Robert. "Incentive Effects of the U.S. Welfare System: A Review." Journal of
Economic Literature 30 (March 1992): 1-61.
Ravallion, Martin and Lorraine Dearden. "Social Security in a ’Moral’ Economy: An
Empirical Analysis for Java." Review of Economics and Statistics 70 (February
1988): 36-44.
Roberts, Russell D. "A Positive Model of Private Charity and Public Transfers." Journal of
Political Economy 92 (June 1984): 136-148.
Schoeni, Robert F. “Private Interhousehold Transfers of Money and Time: New Empirical
Evidence.” Review of Income and Wealth 43 (December 1997): 423-448.
Townsend, Robert M. “Risk and Insurance in Village India.” Econometrica 62 (May 1994):
539-591.
Wilhelm, Mark O. “Bequest Behavior and the Effect of Heirs’ Earnings: Testing the Altruistic
Model of Bequests.” American Economic Review 86 (September 1996): 874-92.
The World Bank. World Development Report. New York: Oxford University Press, 1990.
_____________. The Philippines: An Opening for Sustained Growth. Report no. 11061-PH.
Washington, DC, processed, 1993.
_____________. A Strategy to Fight Poverty. Washington, DC: The World Bank, 1996.
29
Table 1
Selected Characteristics of Filipino Urban Households by Private-Transfer Status
Variable
(1)
Net Transfer
Recipients
Income
Total income before transfers
Proportion with retirement income
Retirement income
Total income after transfers
46,730.30
0.063
862.429
56,796.54
123,720.50
0.078
2,225.170
122,460.44
81,937.29
0.066
1,014.570
81,937.29
59,522.50
0.065
1,031.140
67,246.81
0.175
0.200
0.125
0.221
0.141
0.139
0.137
0.144
0.114
0.203
0.133
0.268
0.138
0.180
0.117
0.227
0.163
0.174
0.166
0.191
0.123
0.219
0.143
0.158
45.376
0.824
0.185
0.291
0.183
0.115
0.789
1.071
5.270
47.072
0.833
0.131
0.385
0.115
0.090
0.650
1.018
5.425
44.715
0.861
0.114
0.312
0.132
0.103
0.763
1.003
5.390
45.482
0.830
0.170
0.304
0.169
0.111
0.771
1.057
5.302
Transfers
Proportion giving net transfers
0.000
Net transfers given (amount)
0.000
Proportion receiving net transfers
1.000
Net transfers received (amount)
10,066.24
Proportion giving gross transfers
0.429
Gross transfers given (amount)
194.22
Proportion receiving gross transfers
1.000
Gross transfers received (amount)
10,260.46
Proportion receiving from abroad
0.264
Transfers received from abroad (amount) 6,518.75
1.000
1,260.060
0.000
0.000
1.000
1,511.57
0.451
251.514
0.005
25.000
0.000
0.000
0.000
0.000
0.008
2.891
0.008
2.891
0.000
0.000
0.110
138.758
0.767
7,724.31
0.441
315.844
0.818
7,901.39
0.203
5,004.90
Education
Some primary or none
Primary graduate
Some secondary
Secondary graduate
Some university
University graduate
Other Characteristics
Age of household head
Married
Female-headed households
Husband and wife both work
Head not employed
No. of children aged 1 or less
No. of children aged 1 to 7
No. of children aged 8 to 15
Household size
Number of cases
(2)
Net Transfer
Donors
6,801
a. Neither a net-transfer recipient nor a net-transfer donor.
976
(3)
Othersa
1,086
(4)
All
Households
8,863
30
Table 2
Selected Characteristics of Filipino Rural Households by Private-Transfer Status
Variable
(1)
Net Transfer
Recipients
(2)
Net Transfer
Donors
Income
Total income before transfers
Proportion with retirement income
Retirement income
Total income after transfers
22,899.29
0.023
220.539
26,711.54
41,539.12
0.021
249.757
40,749.30
26,792.76
0.013
141.505
26,792.76
25,098.24
0.022
218.205
28,255.98
0.438
0.276
0.094
0.106
0.051
0.036
0.424
0.230
0.109
0.107
0.058
0.071
0.426
0.253
0.107
0.130
0.032
0.051
0.436
0.269
0.096
0.108
0.050
0.041
46.402
0.849
0.123
0.271
0.094
0.129
0.896
1.249
5.290
45.119
0.907
0.066
0.270
0.040
0.102
0.850
1.237
5.390
45.218
0.862
0.102
0.259
0.063
0.108
0.862
1.192
5.211
46.188
0.856
0.116
0.271
0.086
0.125
0.889
1.244
5.295
Transfers
Proportion giving net transfers
0.000
Net transfers given (amount)
0.000
Proportion receiving net transfers
1.000
Net transfers received (amount)
3,812.25
Proportion giving gross transfers
0.473
Gross transfers given (amount)
115.663
Proportion receiving gross transfers
1.000
Gross transfers received (amount)
3,927.92
Proportion receiving from abroad
0.126
Transfers received from abroad (amount) 1,835.43
1.000
789.823
0
0
1.000
1,023.99
0.611
234.165
0.009
8.075
0.000
0.000
0.000
0.000
0.013
2.792
0.013
2.792
0.000
0.000
0.104
81.974
0.828
3,157.74
0.497
202.272
0.893
3,278.04
0.105
1,521.15
Number of cases
1,044
Education
Some primary or none
Primary graduate
Some secondary
Secondary graduate
Some university
University graduate
Other Characteristics
Age of household head
Married
Female-headed households
Husband and wife both work
Head not employed
No. of children aged 1 or less
No. of children aged 1 to 7
No. of children aged 8 to 15
Household size
8,332
a. Neither a net-transfer recipient nor a net-transfer donor.
(3)
Othersa
683
(4)
All
Households
10,059
31
Table 3
Transfer Functions
Dependent Variable---Net Transfers Receiveda
Variable
Income
Income Threshold (K)
Income below K
Income above K
No Income
Retirement income
Has retirement income
Education
Primary graduate
Some secondary
Secondary graduate
Some university
University graduate
Other Characteristics
Age of household head
Female-headed households
Married
Married and Female-headed
No. of children aged 1 or less
No. of children aged 1 to 7
No. of children aged 8 to 15
No. of adults
Husband and wife both work
Head not employed
Constant
Observations
R-squared
(1)
(2)
Urban Households
Rural Households
Coefficient
20,080
-0.373
-0.002
11,586
-0.039
-1,225
Standard Error Coefficient
(1967)
(0.075)
(0.006)
(2899)
(0.025)
(896)
10,576
-0.372
-0.028
14,066
-0.109
933
1175
697
3873
6425
6901
(634)
(717)
(645)
(723)
(754)
1,168
1,352
2,502
4,415
5,022
-32
1,265
322
32,131
-66
-384
377
850
-1,399
8,398
-835
(18)
(1005)
(910)
(1364)
(577)
(210)
(160)
(129)
(443)
(595)
(1369)
18.0
689
372
15,836
-43
-43
226
558
-506
5,632
-1,815
8,684
0.24
a. Dependent variable is gross transfers received minus gross transfers given.
Standard Error
(995)
(0.074)
(0.007)
(4686)
(0.056)
(788)
(204)
(296)
(292)
(404)
(494)
(7.7)
(449)
(395)
(733)
(240)
(91)
(65)
(68)
(194)
(333)
(581)
9,857
0.17
32
Table 4
Bootstrap Tests of Polynomial Transfer Function
Against Spline Transfer Function
Polynomial
Order
Urban
Wald Statistic
1
2
3
4
5
6
7
49.8
38.2
28.9
20.0
21.0
15.3
14.3
Rural
P-Valuea
Wald Statistic
0.00
0.00
0.00
0.02
0.03
0.20
0.35
41.4
23.7
14.7
10.3
10.0
11.0
13.4
P-Valuea
0.00
0.00
0.04
0.27
0.42
0.45
0.32
Table 5
Bootstrap Tests of Single Knot Spline
Against Double Knot Spline
Urban
Wald Statistic
0.37
Rural
P-Valuea
0.97
Wald Statistic
2.53
P-Valuea
0.45
a. Percentage of simulated Wald statistics which exceed sample Wald statistic out of 1000 bootstrap replications. Pvalues less than 0.05 indicate rejection at the 5% level.
33
Figure 1.
The Relationship Between Private Transfers and Income
Transfers (T)
Ir’
Ir"’
Recipient Income (Ir)
Ir"